A six-cylinder leveling system and method based on dual tilt sensing and neural network

Abstract

This invention discloses a six-cylinder leveling system and method based on dual tilt sensing and neural networks, belonging to the field of industrial automation leveling technology. The system is configured as follows: a base tilt sensing unit is located below the leveling platform and connected to the platform's mounting base, collecting the tilt angle of the mounting base in real time; the device to be leveled is positioned above the leveling platform, fitting snugly against it; a device tilt sensing unit is connected to the device and installed at the bottom center of gravity of the device, collecting the real-time tilt angle; each servo cylinder is connected to a corresponding cylinder drive unit; the cylinder drive unit collects the absolute positions of the six servo cylinders in real time and receives movement commands from the control unit, driving the six servo cylinders to the target position. This invention can eliminate the initial tilt interference of the mounting base of the leveling platform, does not rely on traditional inverse geometry, avoids the accumulation of errors in relative adjustment, and achieves high leveling accuracy and fast response.

Description

Technical Field

[0001] This invention belongs to the field of industrial automation leveling technology, specifically relating to a six-cylinder leveling system and method based on dual tilt sensing and neural networks. Background Technology

[0002] In high-end manufacturing and scientific research fields such as aerospace precision testing, optical device calibration, ultra-precision machining, and inertial navigation equipment calibration, the levelness of the leveling platform supporting precision equipment directly determines the accuracy of testing data, machining precision, and equipment operational stability. Even a small angular deviation can cause systematic errors. Therefore, extremely high requirements are placed on the static accuracy, dynamic response speed, load-bearing rigidity, and environmental adaptability of the leveling platform. Currently, most conventional small and medium-sized precision leveling platforms adopt a three-point, three-servo electric cylinder independent support structure. Relying on three servo electric cylinders distributed in a triangle as actuators, the platform attitude is adjusted by individually adjusting the extension and retraction of each cylinder. This structure has a certain degree of attitude adjustment flexibility in light-load, small-platform scenarios, but under heavy-load, large-platform, and ultra-high-precision conditions, it suffers from poor load uniformity, insufficient rigidity, and susceptibility to local deformation, making it difficult to meet the leveling requirements of large precision equipment. While the industry uses a six-cylinder support architecture for heavy-load, large-size platforms, which can improve load-bearing rigidity and platform stability, its supporting control and sensing solutions still follow traditional logic and generally have multiple inherent defects, as follows:

[0003] I. Conventional leveling scheme using traditional geometric calculations + single tilt angle sensing:

[0004] This solution is currently the mainstream implementation method for industrial and commercial precision leveling platforms. Regardless of whether it is a three-cylinder or six-cylinder support architecture, the hardware configuration and control logic are highly solidified: only a single dual-axis tilt sensor is set on the upper surface of the leveling platform and in the installation area of ​​the equipment to be leveled to collect the real-time tilt angle of the platform; the control system is based on a PLC or embedded motion controller, with a built-in preset rigid body attitude geometry calculation formula. The X and Y axis tilt deviations collected by the single tilt sensor are substituted into the ideal model to calculate the target extension and retraction of each electric cylinder, and then the electric cylinder is driven to move through PID closed-loop control until the tilt angle of the platform reaches the target.

[0005] The core drawbacks of this solution are extremely prominent: First, it completely ignores the installation tilt state of the leveling platform's mounting base itself, only collecting composite tilt angle data of the platform. This makes it impossible to distinguish whether the tilt deviation originates from an inherent tilt caused by the base's ground / mounting base being uneven, or from platform tilt caused by electric cylinder attitude deviation, uneven load, or mechanical deformation. In practical applications, leveling platforms are often placed directly on non-perfectly level carriers such as workshop floors, vehicle-mounted bases, shipborne bases, and airborne platforms. The initial tilt of the base is superimposed on the platform tilt angle readings throughout the process, resulting in a severe disconnect between the geometric solution model and the actual physical scenario. Second, the traditional geometric formula is an ideal rigid body model, failing to consider nonlinear factors such as electric cylinder assembly gaps, load deformation, mechanical hysteresis, and multi-actuator coupling. With a six-cylinder architecture, the increased number of actuators and intensified coupling effects further amplify the leveling error. This not only makes it difficult to meet the accuracy requirements of ultra-precision scenarios, but also requires manual recalibration after changing the installation site, resulting in extremely poor adaptability. In addition, this scheme relies on PID successive correction, which results in slow leveling response and is prone to overshoot and oscillation. The coordinated control of six electric cylinders is difficult, the steady-state accuracy is low, and the platform is prone to local tilting under heavy load conditions.

[0006] II. Improved leveling scheme assisted by a single neural network:

[0007] To compensate for the nonlinear fitting defects of traditional geometric solutions, some high-end leveling platforms have attempted to introduce neural network algorithms for optimization, forming an improved scheme of single tilt angle sensing + neural network: still using a single platform tilt angle sensor to collect equipment side tilt data, combined with real-time position data of each electric cylinder to build training samples, and using a neural network to fit the mapping relationship between the electric cylinder position and the platform tilt angle, replacing the traditional geometric formula to complete the adjustment calculation.

[0008] However, this type of solution has not overcome the core perception bottleneck. The primary problem is that it does not incorporate the base installation tilt as an independent input into the model training. It simply fits the relationship between the electric cylinder position and a single tilt angle, which is essentially a black-box fitting and cannot isolate the fixed interference caused by the base tilt. This problem is further amplified for the six-cylinder support architecture: on the one hand, the coupling of the six-cylinder collaborative control is stronger, resulting in extremely poor model generalization ability. It can only adapt to a single fixed installation scenario, and the model completely fails after changing the installation base, requiring the collection of massive amounts of samples for retraining; on the other hand, the model training samples do not distinguish between base interference and platform posture deviation, and the learned mapping relationship has systematic errors. Even if the sample size and network complexity are increased, the steady-state error caused by the base tilt cannot be eliminated at its root, and the leveling accuracy bottleneck is difficult to overcome. At the same time, existing solutions mostly use general neural network structures and have not been specifically optimized for the multi-input, strong coupling, and multi-output characteristics of six-cylinder leveling. The inference speed is slow, making it difficult to meet the real-time and synchronous leveling requirements of precision equipment, and it fails to take advantage of the heavy-duty and high-rigidity structural advantages of the six-cylinder system.

[0009] In summary, existing six-cylinder leveling platforms, regardless of whether they employ traditional geometric calculations or single neural network optimization, all suffer from common problems such as limited perception dimensions, failure to consider the tilt of the base mounting reference, disconnect between the model and actual physical working conditions, and poor precision in multi-actuator collaborative control. This results in insufficient leveling accuracy, poor environmental adaptability, slow response speed, and insufficient rigidity under heavy load conditions, making them unsuitable for the all-weather, multi-scenario, and highly stable leveling requirements of ultra-precision scenarios such as large precision equipment, aerospace, and optical inspection. Summary of the Invention

[0010] To solve the above-mentioned technical problems, the present invention adopts the following technical solution:

[0011] A six-cylinder leveling system based on dual tilt sensing and neural network includes: a base tilt sensing unit, six servo electric cylinders, a leveling platform, an equipment tilt sensing unit, the equipment to be leveled, a control unit, and an electric cylinder drive unit.

[0012] The base tilt sensing unit is located below the leveling platform and is rigidly connected to the mounting base of the leveling platform to collect the tilt angle of the mounting base of the leveling platform in real time.

[0013] The equipment to be leveled is positioned above the leveling platform and is flush with the leveling platform;

[0014] The equipment tilt sensing unit is rigidly connected to the equipment to be leveled and is installed at the bottom center of gravity of the equipment to be leveled or on the contact surface between the equipment to be leveled and the leveling platform. It is used to collect the real-time tilt angle of the equipment to be leveled.

[0015] Each servo cylinder is connected to a corresponding cylinder drive unit; the cylinder drive unit is used to collect the absolute position of the six servo cylinders in real time, and at the same time receive the movement command issued by the control unit to drive the six servo cylinders to move synchronously and accurately to the target position.

[0016] The control unit has a built-in multilayer perceptron neural network model that receives the initial tilt angle data of the leveling platform and the real-time tilt angle data of the equipment to be leveled in real time. After rapid inference by the model, it outputs the optimal target absolute position of the six servo electric cylinders and synchronously controls the action of the electric cylinder drive unit to achieve deviation-free coordinated leveling of the six servo electric cylinders.

[0017] A six-cylinder leveling method based on dual tilt sensing and neural network, used in the aforementioned six-cylinder leveling system based on dual tilt sensing and neural network, comprising:

[0018] S1 controls six servo electric cylinders to step-adjust within a preset stroke range, records the tilt angle of the leveling platform mounting base corresponding to the absolute position of each group of electric cylinders and the real-time tilt angle of the equipment to be leveled, and constructs a training sample set;

[0019] S2. Standardize and preprocess the sample set. First, use the absolute position of the servo electric cylinder and the tilt angle of the mounting base of the leveling platform as input features, and the tilt angle of the equipment as the label to train the initial multilayer perceptron neural network model. Then, use the tilt angle of the mounting base of the leveling platform and the real-time tilt angle of the equipment to be leveled as input, and the absolute position of the servo electric cylinder as the label to fine-tune and obtain a multilayer perceptron neural network model adapted to six servo electric cylinders.

[0020] S3, after power-on, the tilt angle of the initial leveling platform mounting base is collected and the benchmark is calibrated through the base tilt sensing unit. During the operation of the equipment, the real-time tilt angle of the equipment to be leveled is collected. If the tilt angle of the equipment to be leveled exceeds the preset tilt angle threshold, the tilt angle of the leveling platform mounting base and the real-time tilt angle of the equipment to be leveled are input into the multilayer perceptron neural network model, and the absolute position of the synchronous target of the six servo electric cylinders is output.

[0021] S4, the electric cylinder drive unit receives the optimal target absolute position sent by the control unit, drives the six servo electric cylinders to move synchronously to the optimal target absolute position, until the tilt angle of the equipment to be leveled is lower than the preset tilt angle threshold, and completes the high-precision closed-loop leveling.

[0022] The present invention has the following beneficial effects:

[0023] This invention constructs a dual-path tilt sensing architecture by adding a base tilt sensing unit and an equipment tilt sensing unit. It collects the initial tilt of the leveling platform mounting base and the real-time tilt of the equipment to be leveled, respectively. This can accurately distinguish the source of tilt error and effectively eliminate the interference of the initial tilt of the leveling platform mounting base. It solves the problems of traditional single tilt angle sensing, such as the inability to identify the reference tilt of the leveling platform mounting base, the disconnect between the model and the actual working conditions, and poor environmental adaptability.

[0024] This invention employs a control unit with a built-in dedicated multilayer perceptron (MLP) neural network model, which directly outputs the absolute position of the six electric cylinders using dual tilt data as input. This achieves the elimination of traditional geometric inverse solution, avoidance of cumulative error in relative adjustment, and deviation-free coordinated control of the six electric cylinders. It solves the problems of poor nonlinear adaptation of geometric solution models, large coupling error of multiple electric cylinders, and low leveling accuracy.

[0025] This invention utilizes a unique training method that pre-trains and then fine-tunes the multilayer perceptron (MLP) neural network model, combined with full-process sample coverage and data standardization preprocessing. This enables the model to achieve fast inference speed, strong generalization ability, and eliminates the need for manual recalibration when changing installation scenarios. It solves the problems of poor generalization, insufficient real-time performance, and weak reusability in multiple scenarios in existing neural network leveling schemes.

[0026] This invention achieves single-time positioning, fast leveling response, no overshoot or oscillation, and high steady-state accuracy by directly driving the absolute position of the electric cylinder and using real-time closed-loop feedback leveling. It solves the problems of slow response, easy oscillation, and difficulty in leveling under heavy load conditions of traditional PID successive correction.

[0027] This invention achieves accurate attitude acquisition, high data consistency, and strong resistance to mechanical deformation and noise interference by arranging the sensor (equipment tilt sensing unit) at the center of gravity of the equipment and averaging multiple samples. It solves the problems of non-standard sensor installation, easy influence of local deformation on detection data, and insufficient leveling reliability. Attached Figure Description

[0028] Figure 1 This is a schematic diagram of the six-cylinder leveling system based on dual tilt sensing and neural network of the present invention, wherein 1-base tilt sensing unit, 2-servo electric cylinder, 3-leveling platform, 4-equipment tilt sensing unit, 5-equipment to be leveled, 6-control unit, and 7-electric cylinder drive unit.

[0029] Figure 2 This is a schematic diagram of the structure of a multilayer perceptron (MLP) neural network model. Detailed Implementation

[0030] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention. Furthermore, the technical features involved in the various embodiments of this invention described below can be combined with each other as long as they do not conflict with each other.

[0031] This invention provides a six-cylinder leveling system and method based on dual tilt sensing and neural networks. The system collects the tilt state of the leveling platform 3 and the equipment to be leveled 5 through dual tilt sensing units. Combined with a specially optimized multilayer perceptron neural network model, the system accurately outputs the optimal absolute position of the six electric cylinders, driving them directly to the target workstation. This achieves heavy-duty, high-precision, high-response, and high-stability planar leveling, fully leveraging the load-bearing advantages of the six-cylinder support structure.

[0032] like Figure 1 As shown, the six-cylinder leveling system based on dual tilt sensing and neural network disclosed in this invention specifically includes: a base tilt sensing unit 1, six servo electric cylinders 2 that are symmetrically distributed and uniformly arranged, a leveling platform 3, an equipment tilt sensing unit 4, an equipment to be leveled 5, a control unit 6, and an electric cylinder drive unit 7.

[0033] The base tilt sensing unit 1 uses a high-precision dual-axis tilt meter, which is installed below the leveling platform 3 and rigidly connected to the leveling platform mounting base. It collects the initial tilt angle of the leveling platform mounting base in real time and is not affected by the table posture or equipment load.

[0034] The leveling device 5 is placed above the leveling platform 3 and is in contact with the leveling platform 3.

[0035] The equipment tilt sensing unit 4 is rigidly connected to the equipment 5 to be leveled and is installed at the bottom center of gravity of the equipment 5, or on the contact surface between the equipment 5 and the leveling platform 3, to collect the real-time tilt angle of the equipment 5. It accurately reflects the actual level status of the equipment 5 to be leveled.

[0036] Electric cylinder drive unit 7: Each servo electric cylinder 2 is connected to one electric cylinder drive unit 7. A total of six servo electric cylinders 2 are symmetrically and evenly distributed under the leveling platform 3 in a six-degree-of-freedom parallel structure, forming a regular hexagonal heavy-duty high-rigidity support structure to ensure that the platform surface of the leveling platform 3 has uniform load-bearing capacity and stable force distribution. The electric cylinder drive unit 7 is used to collect the absolute position of the six servo electric cylinders 2 in real time. , , , , , Simultaneously, it receives movement commands from the control unit 6 and drives the six servo electric cylinders 2 to move synchronously and precisely to the target position, eliminating coordination lag errors.

[0037] The control unit 6 employs an embedded industrial control module, incorporating a pre-trained multilayer perceptron (MLP) neural network model, along with a data preprocessing module, a threshold judgment module, and a drive command distribution module. The data preprocessing module filters, standardizes, and removes outliers from the dual-channel tilt data and electric cylinder position data, ensuring stable and reliable model input. The threshold judgment module compares the tilt angle of the device to be leveled 5 with a preset threshold in real time to determine whether automatic leveling is triggered. The drive command distribution module parses the target absolute positions of the six electric cylinders into drive commands and synchronously distributes them to each electric cylinder drive unit 7, ensuring seamless coordinated action among the six electric cylinders. The control unit 6 receives the initial tilt angle data from the leveling platform 3 and the real-time tilt angle data from the device to be leveled 5 in real time, and outputs the optimal target absolute positions of the six servo electric cylinders 2 after rapid model inference. , , , , , The synchronous control of the electric cylinder drive unit 7 enables the six servo electric cylinders 2 to achieve deviation-free coordinated leveling.

[0038] The six-cylinder leveling method based on dual tilt sensing and neural networks of the present invention is implemented by relying on dual tilt sensing and multilayer perceptron (MLP) neural network model, and includes the following specific steps:

[0039] S1, Sample Collection Stage:

[0040] Calibrate the initial position of the servo cylinder 2: Synchronously adjust the six servo cylinders 2 to the mechanical center position, and record the absolute position of the six servo cylinders 2 at this time as the system reference position.

[0041] Step-adjustable servo cylinder 2: Controls each servo cylinder 2 within its rated stroke range, making fine adjustments in fixed steps to cover reasonable position combinations within the full stroke of the six servo cylinders 2, conforming to actual working conditions.

[0042] Data collection: After each adjustment of the position of servo cylinder 2 and once the system stabilizes, the absolute positions of all six servo cylinders 2 are collected synchronously. , , , , , Adjust the tilt angle of the platform mounting base. The real-time tilt angle of the device to be leveled 5 This constitutes a standardized sample set. Each standardized sample set is obtained by repeatedly collecting samples multiple times (e.g., 3 times) and taking the average value. This effectively eliminates sensor noise interference, and finally constructs a standardized sample set of no less than 500 standardized samples.

[0043] S2, Model Training Phase:

[0044] Sample preprocessing: This includes determining the absolute position of the six servo electric cylinders 2 in the standardized sample set, and the tilt angle of the leveling platform mounting base. Real-time tilt angle of the device to be leveled 5 All scores were standardized using Z-scores (standard scores), with the following formula: ,in, For standardized data, For the original sample data, The sample mean. The standard deviation of the samples is used to eliminate differences in data dimensions and improve the training efficiency and accuracy of multilayer perceptron (MLP) neural network models.

[0045] Model Construction: A multilayer perceptron (MLP) neural network model adapted to the leveling of six servo electric cylinders was built. The structure of the multilayer perceptron (MLP) neural network model is as follows: Figure 2 As shown, the input layer dimension of the multilayer perceptron (MLP) neural network model is 4, and the corresponding input feature is the tilt angle of the mounting base of the leveling platform. The real-time tilt angle of the device to be leveled 5 The output layer has a dimension of 6, corresponding to the absolute positions of the six servo electric cylinders 2. , , , , , The hidden layer consists of two fully connected layers with progressively narrower layer widths. The first fully connected layer has 64 neurons, and the second fully connected layer has 32 neurons. The layer parameters are matched to the multi-dimensional coupling mapping characteristics of the six servo electric cylinders. Each fully connected layer is followed by a ReLU activation function to effectively solve the gradient vanishing problem in nonlinear fitting and accurately adapt to nonlinear working conditions such as electric cylinder mechanical hysteresis, load deformation, and multi-actuator coupling. To address the issues of overfitting and poor generalization in training under the condition of multiple actuators coupled and leveling six servo electric cylinders, L2 (norm 2) regularization constraints are introduced into each fully connected layer, with the regularization coefficient set to 0.001. Additionally, a random deactivation layer is set after each hidden layer (and each fully connected layer), with a dropout rate of 0.1, randomly weakening the coupling relationships of some neurons and improving the robustness and scene adaptability of the multilayer perceptron neural network model. The output layer has no activation function to meet the requirement of continuous numerical output of the absolute position of the electric cylinders. The training of the multilayer perceptron neural network model uses the mean squared error (MSE) loss function to accurately quantify the prediction deviation of the electric cylinder position.

[0046] Model Training: The standardized sample set is divided into training and test sets according to a certain ratio. The Adam (Adaptive Moment Estimator) optimizer is used, and parameters such as learning rate and number of iterations are set (e.g., learning rate = 0.001, number of iterations = 1000, batch size = 32) until the test set loss value is ≤1e-6 to ensure the inference accuracy of the multilayer perceptron (MLP) neural network model. First, the absolute position of the servo cylinder 2 and the tilt angle of the leveling platform 3 are used as inputs, and the real-time tilt angle of the device to be leveled 5 is used as the label to complete the pre-training. Then, the tilt angle of the mounting base of the leveling platform and the real-time tilt angle of the device to be leveled 5 are used as inputs, and the absolute position of the servo cylinder 2 is used as the label to fine-tune the final multilayer perceptron (MLP) neural network model, so that the multilayer perceptron (MLP) neural network model can directly fit the mapping relationship between the dual tilt data and the target position of the electric cylinder.

[0047] Model Deployment: The trained multilayer perceptron (MLP) neural network model is embedded into the embedded industrial control module of the control unit 6 to ensure real-time inference speed.

[0048] S3, Real-time Leveling Stage:

[0049] S3.1, Start-up calibration of the reference tilt angle of leveling platform 3: After the system is powered on, the tilt angle of the mounting base of the leveling platform is collected as the fixed reference value for this operation, and the interference of the initial tilt of the mounting base of the leveling platform is eliminated.

[0050] S3.2, Real-time attitude monitoring: During equipment operation, the equipment tilt sensing unit 4 frequently collects the real-time tilt angle of the equipment to be leveled 5. If the tilt angle of the equipment to be leveled 5 exceeds the preset tilt angle threshold... This immediately triggers the automatic leveling command. Preset tilt angle threshold. It can be adaptively adjusted according to the accuracy requirements of the equipment.

[0051] S3.3, Rapid Model Inference: Input the tilt angle of the current leveling platform 3 and the real-time tilt angle of the device to be leveled 5 into the solidified multilayer perceptron (MLP) neural network model, and quickly output the synchronous optimal target absolute position of the six servo electric cylinders 2.

[0052] S3.4, Synchronous drive of electric cylinders: The electric cylinder drive unit 7 receives the optimal target absolute position sent by the control unit 6 and directly drives the six servo electric cylinders 2 to move synchronously from the current absolute position to the optimal target absolute position, without the need for step-by-step adjustment, thus eliminating accumulated errors.

[0053] S4, Closed-loop feedback stage:

[0054] After a single leveling action is completed, a delay is set for stabilization. The equipment tilt sensing unit 4 immediately collects the real-time tilt angle of the equipment 5 to be leveled again. If it still exceeds the preset tilt angle threshold... The rapid model inference in S3.3 and the synchronous electric cylinder drive in S3.4 are repeated to form a closed-loop feedback. During the leveling cycle, the tilt attitude data of the device to be leveled 5 is collected every predetermined time (e.g., 0.1s) until the real-time tilt angle of the device to be leveled 5 is lower than the preset tilt angle threshold. The leveling process is now complete, ensuring steady-state accuracy.

[0055] In this embodiment, the equipment tilt sensing unit 4 is fixed at the center of gravity of the equipment 5 to be leveled, away from the cantilever and moving parts, to eliminate detection errors caused by local deformation of the equipment.

[0056] Six servo electric cylinders 2 are symmetrically and evenly distributed below the leveling platform 3 in a six-degree-of-freedom parallel structure, forming a regular hexagonal heavy-duty high-rigidity support structure. The electric cylinder drive unit 7 supports synchronous closed-loop control and uses absolute position commands throughout the process, eliminating the need to calculate and accumulate relative adjustment amounts, thus avoiding leveling deviations caused by accumulated errors and transmission lag from the root.

[0057] The multilayer perceptron (MLP) neural network model built into the control unit 6 outputs the absolute positions of the six servo electric cylinders 2. During the training phase, it fully learns the fixed mapping relationship between the tilt angle of the leveling platform mounting base, the tilt angle of the device to be leveled 5, and the electric cylinder position. In actual operation, after the S3.1 power-on calibration of the tilt angle of the leveling platform 3, it can quickly eliminate the interference of the installation environment. After the model inferences quickly in S3.3, it directly sends the optimal target absolute position to the electric cylinder drive unit 7, driving the servo electric cylinder 2 to be positioned in one go. The leveling speed is fast and the steady-state accuracy is high (leveling accuracy reaches 0.005°, response time <2 seconds). There is no need for manual recalibration when changing the installation site. The multilayer perceptron (MLP) neural network model has strong adaptive ability and is suitable for heavy-duty precision leveling scenarios with various non-ideal bases.

[0058] The above description is merely an embodiment of the present invention and does not limit the scope of the invention. Any equivalent structural or procedural transformations made based on the description and drawings of this invention, or direct or indirect applications in other related system fields, are similarly included within the protection scope of this invention. Contents not described in detail in this specification are prior art known to those skilled in the art.

Claims

1. A six-cylinder leveling method based on dual tilt sensing and neural network, used in a six-cylinder leveling system, characterized in that, The six-cylinder leveling system includes: a base tilt sensing unit, six servo electric cylinders, a leveling platform, an equipment tilt sensing unit, the equipment to be leveled, a control unit, and an electric cylinder drive unit. S1 controls six servo electric cylinders to step-adjust within a preset stroke range, records the tilt angle of the leveling platform mounting base corresponding to the absolute position of each group of electric cylinders and the real-time tilt angle of the equipment to be leveled, and constructs a training sample set; The base tilt sensing unit, located below the leveling platform, is rigidly connected to the mounting base of the leveling platform and is used to collect the tilt angle of the mounting base of the leveling platform in real time. The equipment to be leveled is positioned above and in contact with the leveling platform. The equipment tilt sensing unit, which is rigidly connected to the equipment to be leveled, is installed at the bottom center of gravity of the equipment to be leveled or on the contact surface between the equipment to be leveled and the leveling platform, and is used to collect the real-time tilt angle of the equipment to be leveled. Each of the six servo cylinders is connected to a corresponding cylinder drive unit, which is used to collect the absolute position of the six servo cylinders in real time, and at the same time receive the movement command issued by the control unit to drive the six servo cylinders to move synchronously and accurately to the target position; S2, the sample set is standardized and preprocessed. First, the absolute position of the servo cylinders and the tilt angle of the mounting base of the leveling platform are used as input features, and the tilt angle of the equipment is used as the label to train the initial multilayer perceptron neural network model; then, the tilt angle of the mounting base of the leveling platform and the real-time tilt angle of the equipment to be leveled are used as input, and the absolute position of the servo cylinders is used as the label to fine-tune and obtain a multilayer perceptron neural network model adapted to the six servo cylinders; S3, after power-on, the tilt angle of the initial leveling platform mounting base is collected and the benchmark is calibrated through the base tilt sensing unit. During the operation of the equipment, the real-time tilt angle of the equipment to be leveled is collected. If the tilt angle of the equipment to be leveled exceeds the preset tilt angle threshold, the tilt angle of the leveling platform mounting base and the real-time tilt angle of the equipment to be leveled are input into the multilayer perceptron neural network model, and the absolute position of the synchronous target of the six servo electric cylinders is output. S4, the electric cylinder drive unit receives the optimal target absolute position sent by the control unit, drives the six servo electric cylinders to move synchronously to the optimal target absolute position, until the tilt angle of the equipment to be leveled is lower than the preset tilt angle threshold, and completes the high-precision closed-loop leveling.

2. The six-cylinder leveling method based on dual tilt sensing and neural network according to claim 1, characterized in that, In S1, the control unit has a built-in multilayer perceptron neural network model that receives the initial tilt angle data of the leveling platform and the real-time tilt angle data of the equipment to be leveled in real time. After the model quickly infers, it outputs the optimal target absolute position of the six servo electric cylinders and synchronously controls the action of the electric cylinder drive unit to achieve deviation-free coordinated leveling of the six servo electric cylinders.

3. The six-cylinder leveling method based on dual tilt sensing and neural network according to claim 1, characterized in that, In S1, the base tilt sensing unit uses a high-precision dual-axis tilt meter.

4. The six-cylinder leveling method based on dual tilt sensing and neural network according to claim 1, characterized in that, In S1, six servo electric cylinders are symmetrically and evenly distributed below the leveling platform in a six-degree-of-freedom parallel structure, forming a regular hexagonal heavy-duty high-rigidity support structure.

5. The six-cylinder leveling method based on dual tilt sensing and neural network according to claim 1, characterized in that, In S1, the input layer of the multilayer perceptron neural network model has a dimension of 4, corresponding to the tilt angle of the mounting base of the leveling platform and the real-time tilt angle of the equipment to be leveled; the output layer has a dimension of 6, corresponding to the absolute positions of the six servo cylinders; the hidden layer has at least two fully connected layers, with the number of neurons gradually reduced, and each layer is followed by a ReLU activation function; the output layer has no activation function.

6. The six-cylinder leveling method based on dual tilt sensing and neural network according to claim 1, characterized in that, In S1, the mean squared error loss function is used for training the multilayer perceptron neural network model.

7. The six-cylinder leveling method based on dual tilt sensing and neural network according to claim 1, characterized in that, S1 include: Calibrate the initial position of the servo cylinders: Synchronously adjust the six servo cylinders to the mechanical center position, and record the absolute position of the six servo cylinders at this time as the system reference position; Stepping adjustment servo cylinder: controls each servo cylinder to make fine adjustments in fixed steps within its rated stroke range, covering the reasonable position combination within the full stroke of the six servo cylinders; Data collection: After each adjustment of the position of the servo cylinder, once the six-cylinder leveling system based on dual tilt sensing and neural network has stabilized, the absolute position of the six servo cylinders, the tilt angle of the leveling platform mounting base, and the real-time tilt angle of the equipment to be leveled are collected simultaneously to form a set of standardized samples. Each set of standardized samples is collected repeatedly and the average value is taken to obtain a standardized sample set of no less than 500 sets of standardized samples.

8. The six-cylinder leveling method based on dual tilt sensing and neural network according to claim 1, characterized in that, S2 include: The absolute positions of the six servo electric cylinders, the tilt angle of the mounting base of the leveling platform, and the real-time tilt angle of the equipment to be leveled in the standardized sample set are all processed using Z-score standardization. A multilayer perceptron neural network model adapted to leveling with six servo electric cylinders was constructed. The input layer of the multilayer perceptron neural network model has a dimension of 4, corresponding to the tilt angle of the mounting base of the leveling platform and the real-time tilt angle of the equipment to be leveled. The output layer has a dimension of 6, corresponding to the absolute position of the six servo electric cylinders as the output. Two fully connected layers are set in the hidden layer, and a ReLU activation function is configured after each fully connected layer. No activation function is set in the output layer. The standardized sample set is divided into a training set and a test set according to a certain ratio. An adaptive moment estimation optimizer is used, and the learning rate and number of iterations are set until the test set loss value is ≤1e-6. First, the absolute position of the servo electric cylinder and the tilt angle of the leveling platform mounting base are used as inputs, and the real-time tilt angle of the equipment to be leveled is used as the label to complete the pre-training. Then, the tilt angle of the leveling platform mounting base and the real-time tilt angle of the equipment to be leveled are used as inputs, and the absolute position of the servo electric cylinder is used as the label to perform fine-tuning, so as to obtain the final multilayer perceptron neural network model that can directly infer the target workstation. The trained multilayer perceptron neural network model is embedded into the embedded industrial control module of the control unit.

9. The six-cylinder leveling method based on dual tilt sensing and neural network according to claim 1, characterized in that, S3 include: S3.1, Start-up calibration of the reference tilt angle of the leveling platform: After the system is powered on, the tilt angle of the mounting base of the leveling platform is collected as the fixed reference value for this operation, and the initial tilt interference of the mounting base of the leveling platform is removed. S3.2, Real-time attitude monitoring: During equipment operation, the equipment tilt sensing unit collects the real-time tilt angle of the equipment to be leveled at high frequency. If the tilt angle of the equipment to be leveled exceeds the preset tilt angle threshold, an automatic leveling command is immediately triggered. S3.3, Rapid Model Inference: Input the tilt angle of the current leveling platform and the real-time tilt angle of the equipment to be leveled into the solidified multilayer perceptron neural network model, and quickly output the synchronous optimal target absolute position of the six servo electric cylinders. S3.4, Synchronous Electric Cylinder Drive: The electric cylinder drive unit receives the optimal target absolute position sent by the control unit and directly drives the six servo electric cylinders to move synchronously from the current absolute position to the optimal target absolute position.

10. The six-cylinder leveling method based on dual tilt sensing and neural network according to claim 1, characterized in that, S4 includes: after a single leveling action is completed, a delay is set for stabilization. The equipment tilt sensing unit immediately collects the real-time tilt angle of the equipment to be leveled again. If it still exceeds the preset tilt angle threshold, the operation of rapid model inference in S3.3 and synchronous electric cylinder drive in S3.4 is repeated to form a closed-loop feedback. The tilt attitude data of the equipment to be leveled is collected once every predetermined time during the leveling cycle until the real-time tilt angle of the equipment to be leveled is lower than the preset tilt angle threshold, at which point the leveling ends, ensuring steady-state accuracy.

Images